New opportunity

Gen AI Engineer

EXL Talent Acquisition Team · Gurugram, India

About this role

The Gen AI Engineer will manage generative AI pipelines and workflows, extract insights from unstructured data, and build scalable AI solutions. The position is located in Gurugram, Haryana, India.

Responsibilities

Manage production pipelines for GenAI models, debug defects, and assess root causes.
Build and orchestrate LLM-based workflows using frameworks such as LangChain, including prompt engineering and pipeline design.
Implement retrieval-augmented generation (RAG) architectures using vector databases such as Pinecone for semantic search and contextual retrieval.
Develop document ingestion and extraction workflows for unstructured PDFs, scans, and forms using tools such as AWS Extract and GenAI-based techniques.
Work with engineering, product, and business stakeholders to define data requirements, evaluation metrics, and success criteria. Present findings to technical and nontechnical audiences through visualizations, reports, and presentations.
Keep current with Generative AI, LLM, data science, and analytics practices, and mentor junior team members.

Qualifications

Bachelor’s or Master’s degree in Data Science, Computer Science, AI/ML, Statistics, Mathematics, or a related field.
3–6 years of experience in a data science, applied ML, or GenAI role, with a strong project portfolio.
Hands-on experience with scikit-learn, TensorFlow, or PyTorch; practical experience with LLMs, GenAI frameworks, LangChain, and prompt-driven workflows; and a strong understanding of RAG patterns, vector embeddings, and vector databases such as Pinecone.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Generative AIMachine LearningLarge Language ModelsLangChainPrompt EngineeringRetrieval-Augmented GenerationVector EmbeddingsVector DatabasesPineconeSemantic SearchDocument IngestionDocument ExtractionAWS Extractscikit-learnTensorFlowPyTorchData ScienceData AnalyticsData VisualizationPipeline OrchestrationRoot Cause AnalysisMentoring

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Production reliability
Manage GenAI model pipelines, debug defects, and provide root-cause assessments. The role calls for operational judgment, not just prototype development.
LLM workflows and retrieval
Build and orchestrate prompt-driven workflows using frameworks such as LangChain, and implement RAG with embeddings and vector databases such as Pinecone for semantic search and contextual retrieval.
Unstructured document processing
Develop ingestion and extraction workflows for PDFs, scans, and forms using tools such as AWS Extract and GenAI-based techniques.
Evaluation and stakeholder communication
Work with engineering, product, and business stakeholders to define data requirements, evaluation metrics, and success criteria; present findings to technical and nontechnical audiences.
Technical development and mentoring
Stay current with GenAI, LLM, data science, and analytics practices while mentoring junior team members.

Likely challenges to think through

Tracing defects across a pipeline
Preparation hypothesis: a faulty answer could originate in document extraction, retrieval, prompting, or orchestration, making root cause harder to isolate. Prepare a structured debugging example that tests each stage and explains how the cause was verified.
Balancing retrieval quality and scalability
Preparation hypothesis: improving contextual retrieval may require trade-offs among document processing, embedding choices, search relevance, and pipeline complexity. Prepare to explain how you would evaluate RAG outputs and justify design choices against defined success criteria.
Turning varied goals into measurable results
Preparation hypothesis: engineering, product, and business stakeholders may emphasize different outcomes. Prepare an example of translating data requirements into evaluation metrics, then communicating the findings clearly to both technical and nontechnical audiences.

Practice your answers

Technical practice

How would you investigate a defect in a production GenAI pipeline?

What to coverThe role includes debugging production pipelines and assessing root causes. Suggested preparation: Explain how you would locate the failing stage, distinguish data, retrieval, prompt, and model issues, and verify a fix.

How would you design and orchestrate an LLM-based workflow?

What to coverThe posting names LangChain, pipeline design, and workflow orchestration. Suggested preparation: Describe component boundaries, data passed between stages, and how you would make failures observable and recoverable.

How would you improve a prompt-driven workflow that produces inconsistent outputs?

What to coverPrompt engineering is an explicit responsibility. Suggested preparation: Discuss instructions, input context, output constraints, and test cases; explain how you would tell a prompt problem from an upstream data problem.

How would you design a RAG workflow for contextual answers?

What to coverThe role calls for RAG architectures supporting semantic search and contextual retrieval. Suggested preparation: Walk through document preparation, retrieval, context selection, and answer generation, including what you would do when retrieved context is insufficient.

What choices matter when using embeddings and a vector database for retrieval?

What to coverThe posting specifically names vector embeddings and databases such as Pinecone. Suggested preparation: Explain how embedding quality and indexing choices affect relevance, and how you would test retrieval before judging generated answers.

How would you ingest PDFs, scans, and forms for downstream AI use?

What to coverThe role includes document ingestion and extraction from all three document types. Suggested preparation: Describe how you would handle differing document structure and extraction errors, then check whether the resulting content is usable downstream.

How would you assess a GenAI-based document extraction approach?

What to coverThe posting mentions GenAI-based extraction techniques and a tool it calls AWS Extract. Suggested preparation: Explain how you would compare extracted fields with reference documents, identify recurring errors, and weigh quality against workflow complexity.

How would you define success criteria and evaluation metrics for a GenAI solution?

What to coverThe role involves defining data requirements, metrics, and success criteria with engineering, product, and business stakeholders. Suggested preparation: Distinguish retrieval, extraction, and end-to-end answer quality, and explain how the intended use would determine which measures matter.

What would you examine before scaling a GenAI workflow?

What to coverThe posting describes building scalable AI solutions and managing production pipelines. Suggested preparation: Discuss where processing time, throughput, and failure handling could constrain a workflow, and how you would evaluate trade-offs without assuming a particular deployment setup.

How would you select an ML framework for a relevant component?

What to coverThe qualifications name hands-on experience with scikit-learn, TensorFlow, or PyTorch. Suggested preparation: Compare their suitability for a concrete ML task, including development effort and evaluation needs, without assuming the role requires all three.

Behavioural practice

Tell me about a time you took ownership of a difficult production defect.

What to coverProduction debugging and root-cause assessment are stated duties. Suggested preparation: Use a real situation, your investigative actions, and the result; clarify what evidence supported your conclusion.

Describe a time you aligned stakeholders on what an AI or data solution needed to achieve.

What to coverThe role involves working with engineering, product, and business stakeholders on requirements and success criteria. Suggested preparation: Give a real situation, how you reconciled differing priorities, and the resulting agreement or decision.

Tell me about a time you explained technical findings to a nontechnical audience.

What to coverThe posting calls for communicating insights through visualizations, reports, and presentations. Suggested preparation: Describe the situation, how you adapted the explanation, and what the audience understood or decided as a result.

Describe a time you helped a junior colleague develop technically.

What to coverMentoring junior team members is an explicit responsibility. Suggested preparation: Give a real example of the colleague's need, the actions you took, and an observable result without overstating your contribution.

Tell me about a time you evaluated a new AI or analytics practice before adopting it.

What to coverThe role asks for staying current with GenAI, LLMs, data science, and analytics practices. Suggested preparation: Explain the situation, how you tested its relevance against an existing approach, and the outcome, including if you chose not to adopt it.

Sharpen your knowledge

Topics to prepare

Production pipeline failure analysis
Review how to distinguish model, prompt, retrieval, and input-data failures. Suggested exercise: trace a hypothetical incorrect answer through those stages and explain how you would identify its root cause. The posting calls for debugging production GenAI pipelines.
LLM workflow and prompt design
Review how prompts and workflow stages affect outputs. Suggested exercise: diagram a multi-step LLM workflow and explain where you would test prompt changes. The role involves prompt engineering and pipeline design with frameworks such as LangChain.
RAG and semantic retrieval
Review embeddings, vector search, contextual retrieval, and how retrieved context reaches an LLM. Suggested exercise: explain why a mock question retrieves an irrelevant passage and propose a way to test the cause. The posting explicitly calls for RAG and vector-database understanding.
Unstructured document extraction
Review the different extraction challenges posed by PDFs, scans, and forms. Suggested exercise: outline how you would check extracted fields against a small set of source documents. These document types and GenAI-based extraction are named in the role.
Evaluation and success criteria
Review how data requirements, evaluation metrics, and business success criteria differ. Suggested exercise: define a small set of measures for a hypothetical document-question-answering workflow and explain them to a nontechnical stakeholder. The posting requires cross-functional metric definition and communication.
Applied ML framework fundamentals
Refresh the training, validation, and inference concepts you use with scikit-learn, TensorFlow, or PyTorch. Suggested exercise: explain how you would check an example model's errors before using its output in a workflow. Hands-on experience with these frameworks is listed in the qualifications.

Skills to revise

Orchestrate and test an LLM workflow
Suggested exercise: build a small LangChain workflow with two prompt steps, run several test inputs, and record how a prompt change alters results. This practices the LLM orchestration and prompt-driven workflows named in the posting.
Build a small retrieval path
Suggested exercise: embed a short sample document set, retrieve passages for test questions using a vector database such as Pinecone if available, and inspect whether the passages support the answers. This practices the advertised RAG and semantic-search work.
Process and check documents
Suggested exercise: ingest sample PDFs, scans, and forms; extract selected fields using an available method; and compare results with the originals. If available, try the posting's example tool, AWS Extract. This mirrors its ingestion and extraction responsibility.
Debug a pipeline defect
Suggested exercise: introduce a bad input or retrieval result into a sample workflow, trace the resulting output, and write a brief root-cause assessment. The role calls for debugging defects in production GenAI pipelines.
Apply one listed ML framework
Suggested exercise: use scikit-learn, TensorFlow, or PyTorch to run and evaluate a small example model, then explain its errors and limitations. The qualification asks for hands-on experience with these frameworks, without requiring all three.
Present evaluation findings
Suggested exercise: turn results from a sample workflow into one visualization and a short explanation tailored separately to engineering and business audiences. The posting calls for reports, presentations, and communication across technical and nontechnical stakeholders.

Your preparation checklist

Choose a project walkthrough
The posting asks for a strong project portfolio and hands-on ML experience. Prepare a walkthrough of an actual data science, applied ML, or GenAI project, identifying your contribution, results, and any scikit-learn, TensorFlow, or PyTorch use.
Diagram an LLM workflow
The role includes LangChain, prompt engineering, and pipeline design. Prepare a diagram of an LLM workflow you built, or a clearly labeled practice design, showing orchestration, prompt choices, and how you would test changes.
Work through a RAG design exercise
Sketch how documents would become embeddings, enter a vector database such as Pinecone, and support semantic search and contextual retrieval. Explain how you would check whether retrieved content supports the answer.
Map a document extraction pipeline
The posting mentions PDFs, scans, forms, AWS Extract, and GenAI-based extraction. Prepare a sample ingestion-to-extraction flow and identify how you would handle poor-quality inputs and verify extracted fields.
Prepare a production defect case
The role calls for debugging GenAI production pipelines and assessing root causes. Rehearse an actual incident if available, or label it hypothetical; show the symptoms, investigation, root-cause reasoning, fix, and checks afterward.
Draft an evaluation brief
The role involves defining data requirements, metrics, and success criteria with stakeholders. Prepare a one-page example with proposed measures, a simple visual or report outline, and separate explanations for technical and nontechnical audiences.
Prepare collaboration and mentoring questions
Bring a concrete example of mentoring or knowledge-sharing if applicable. Ask how engineering, product, and business teams agree on evaluation criteria, and what production-pipeline and mentoring responsibilities this role would own.

Availability last checked: 4 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…